Regular updates on the latest VC-backed AI startups. Follow along to stay informed!
Flower, a federated learning platform, raised $3.6M in Pre-Seed funding. First Spark Ventures led the round, with participation from Hugging Face CEO Clem Delangue, Factorial Capital, Betaworks, and Pioneer Fund.
Problem to be Solved
The reliance on public data - mostly web data - to train AI is holding back the AI field. Distributed data that’s trapped on devices like phones or in organizational silos like business units within enterprises is out of reach for AI today.
How They Use AI
Flower provides a platform and framework for conducting federated learning. A typical approach to machine learning is to gather data to a central server and use that data to train a model. However, privacy concerns and technical limitations may prevent data from being uploaded and aggregated. Enter federated learning: where instead of sending data to the model, you send copies of the model to the data, compute updates to the model weights, and then recentralize the model and merge the distributed updates.
Business Model
As an open-source framework, Flower does not directly generate revenue. The project could potentially monetize by offering hosted solutions or paid features or services. Flower’s seen impressive uptake over the past several months, with its community of developers growing to just over 2,300, and developers from Fortune 500 companies and academic institutions like Porsche, Bosch, Samsung, Banking Circle, Nokia, Stanford, Oxford, MIT, and Harvard using the platform.
Dropzone AI, the developer of an autonomous agent for investigating security alerts, raised $3.5M in Seed Round funding. The round was led by Decibel Partners, and joined by Pioneer Square Ventures Fund.
Problem to be Solved
Investigating security alerts is tedious and time-consuming - ask any CISO about alert fatigue from their many EDR, firewall, and cloud security solutions. Plus, human security analysts can’t keep up with automated cyber-attacks.
How They Use AI
Dropzone AI uses LLMs to mimic expert security analysts' thought processes and techniques. The agent can process and investigate every security alert from various sources and produce detailed reports and recommendations for human analysts. LLM agents work by recursively following a prompting framework until they achieve an end state. Dropzone likely has a prompting framework that is triggered (API that is called) each time a security alert arrives.
Business Model
TBD. The website currently only offers options to test drive the product or join the waitlist. It will be interesting to see if Dropzone uses a basic SaaS model or incorporates usage-based pricing to cover the LLM API billing.
Weights & Biases, an ML development platform, raised $50M in Series D funding. The round, led by ex-GitHub CEO Nat Friedman and former Y Combinator partner Daniel Gross alongside existing investors Coatue, Insight Partners, Felicis, Bond, BloombergBeta, and Sapphire, values the company at $1.25B and brings the total raised to $250M.
Problem to be Solved
ML practitioners don’t have a great system of record for their experiments and often end up using spreadsheets and screenshots. Weights & Biases supports an ML development life cycle's testing, security, and reliability workflows. Their customer base ranges from startups to enterprises to academic researchers.
How They Use AI
Weights & Biases is a leading experiment logging platform for ML development. It offers an organized approach for logging experimental data such as training loss, validation accuracy, as well as training hyperparameters (non-learned parameters set by the user; e.g., learning rate). Detailed and organized logs accelerate development and enable reproducibility.
Business Model
The company makes money by charging SaaS subscription fees that scale with the number of users, number of “tracked hours” (how long it takes to train a model), and level of support services. Weights & Biases’ solution is integrated into over 20,000 open-source repositories and has 1,000 paying customers including well-funded generative AI model builders like OpenAI, Anthropic, and Hugging Face.